v21: CUDA Error: no kernel image is available for execution on the device (err_no=209)
2080Ti, latest drivers 419.35
v20: works fine!
Any suggestions?
Sorry guys, v21.0 does not support 20 series cards for now. |
|
|
|
Can I use this miner on C31 for GTX 1080ti/win 10?
Yes, it works. Have you got 1.4+ speed? Yes, as I mentioned above, however, the reference performance (1.45g/s cuckatoo31 on 1080ti) was tested under Linux system. Miner hashrate may vary under windows, due to some issue that I am trying to find out. |
|
|
|
Can I use this miner on C31 for GTX 1080ti/win 10?
Yes, it works. |
|
|
|
Thank you!
If it is any help I noticed huge power draw fluctuations on the most affected cards - i don't use power limiter, only undervolting, and loads goes to to 27w sometimes and then get back to 65w. Dunno if it is useful info for debugging.
Hopefully it is not related to windows being 1809, but so far I see it as biggest difference of the affected rig.
 Every bit of information helps. Thanks. |
|
|
|
some gpus on 1 of my rigs have bad perofrmance, on my other rigs i got from 2.7 to 2.9-3.0, which is equal to linux perf. + you can see that even "better" gpus after nr. 5 get 2.5 g/s instead of 2.9 g/s that is min for all the other gpus on all the other rigs and firs 5 gpus on this rig. I'm running conservative OC and UV. CPU load is unusually low for grin29. IMO. Differences of the affected rig: windows 1809, h110 motherboards, others re window 1703 (i think) and biostar b250 (i think) motherboards I think that in previous nbminer versions I would get 2.7 on all gpus on that rig, but i'm not sure, i will try nbminer before 20 or so to verify. Any idea?  It is weired what happened. There are other users who reported similar situations like this. The reference performances table are tested under linux. I will check the compatibility issue on Windows. Hopes to fix this soon. |
|
|
|
Hi
asrock H110 Pro BTC
5xRTX 2080
BTM
Windows 10 -crash seldom and restart computer
Windows 8.1 only crash
so install windows 7 - drivers have been installed well, windows sees graphics cards but miners do not detect cards and fans are spinning all the time.
What's the problem?
Never heard of such things happened. Try new version? |
|
|
|
Hi my titan Vs crash.. please help
v20.0 does not support ttv v21.0 supports, but not using its tensorcores. No future plans for supporting tensorcores on volta cards. |
|
|
|
v21.0 upadted  |
|
|
|
Good job Thanks Please check, crash when try to reconnect to the pool (C31, btc.com) WARN - #033[31mNo response from mining pool, disconnected.
INFO - Logout.
No response from mining pool, disconnected.
WARN - #033[33mQObject::disconnect: Unexpected null parameter
kernel: [26221.903338] nbminer[4829]: segfault at 8 ip 0000000000543b3b sp 00007ffd89978990 error 4
#033[0m!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Mining program unexpected exit.
Reason: Process crashed
Restart miner after 10 secs ...
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Thanks for detailed report. Will be fixed in the next version. |
|
|
|
THX.
if I don't know which gpu will fail, can I select all cards but only run those normal cards?
the thread that control the fail cards can sleep and restart every 5 minutes.
BTW, i find the program detect the cards in a different order with the nvidia-smi.
this causes some trouble for selecting device.
for example, I have an nvidia-smi result as below,
0 GeForce GTX 1080
1 GeForce GTX 1070
2 GeForce GTX 1080
3 GeForce GTX 1070
but nbminer detect the device# as below,
0 GeForce GTX 1080
1 GeForce GTX 1080
2 GeForce GTX 1070
3 GeForce GTX 1070
Currently NBMiner do not has the option to ignore failed cards. NBMiner uses GPU order in CUDA runtime instead of Driver runtime, these two can be different in some rigs. Thanks to @cudapop. I'm taking this into future change plans. |
|
|
|
program crashes when some gpus fail to start.
is it possible to fix this?
Use option '-d 0,1' to select the first 2 cards, for example. |
|
|
|
Hey Nebutech, great work. What is the minimum memory for C31 with your miner? Does 8gb card work on windows 10? or is it only 11gb? Also what is the minimum and maximum parameters for cpu intensity?
8GB cards only work under win7 and linux. --cuckoo-intensity <intensity> Set intensity of cuckoo, cuckaroo, cuckatoo, [1, 12]. Set to 0 means autumatically adapt. Default: 0. |
|
|
|
Testing v14.0 with an RTX 2060, getting a memory access error: [08:26:00] INFO - ============= Device Information =============
[08:26:00] INFO - ID 4: GeForce RTX 2060 5904 MB, CC 75
[08:26:00] INFO - ==============================================
[08:26:00] INFO - NVML initialized.
[08:26:03] INFO - tensority - Logging in to stratum-btm.antpool.com:6666 ...
[08:26:04] INFO - tensority - New job from stratum-btm.antpool.com, ID: 3548867, DIFF: 16.38K
[08:26:04] INFO - tensority - Login succeeded.
[08:26:04] INFO - API: 0.0.0.0:22333
[08:26:04] INFO - API server started.
[08:26:06] INFO - Worker thread started on device 4.
[08:26:06] INFO - tensority - Device 4 seed cache updated.
CUDA Error: an illegal memory access was encountered (err_no=77)
[08:26:06] ERROR - Device 4 exception, exit ...
[08:26:06] FATAL - QThread: Destroyed while thread is still running
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Mining program unexpected exit.
Reason: Process crashed
Restart miner after 10 secs ...
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Works fine with GTX 1060 cards, but won't work on RTX 2060 cards. Try the new v20.0 version to see if it fixes your problem. |
|
|
|
I have problem with Bytom mining performance. My rig used to mine with 4x1080Ti. Performance was ok, almost 5K hash per card. Now I changed 2 cards and have 2x2080 and 2x1080Ti in the same rig. The 2080 performance is as expected, but sadly 1080Ti performance halved! I tried to run both pairs of cards as separate threads, but it's the same. Any help on this?  There is a confirmed issue that mining BTM with mixed 10 series and 20 series cards, which would cause the 10 series cards has much lower hashrate than previous versions. Will be fixed in next version.  |
|
|
|
Volta's tensor cores don't support the INT8 mode supported by Turing's tensor cores, and BTM's tensority algo uses 8-bit matrix multiplication.
That being said, the @NebuTech devs could make a version specifically for Volta's tensor cores which used the FP16 matrix multiply mode supported by Volta, since you can encode INT8 numbers in FP16 format with no loss of precision.
yes, but there are other limitation other than operator type. So do you have plans to add Volta GPUs support? It's very interesting how much hashrate Volta could be. Could it be profitable like RTX GPUs or it's just waste of money No plans to support Volta GPUs. The price of Volta GPUs is higher that Turing, and even if the tensor core on Volta can be used to mine BTM, the theoretical peak hashrate is still only half of which on Turing's. |
|
|
|
i hope this isnt dead
It's not dead. We have several days off for Chinese new year. New versions will come in few days. |
|
|
|
Dear developers do you plan to use rtx tensor cores to increase hashrate on other algorithms?
Tensor cores can only be used on matrix based algorithms. There is no algorithm that I know for now other than tensority which has this attribute. |
|
|
|
Would the Titan RTX work for BTM?
also really wish we could open up volta tensors
Titan RTX should get a hashrate above 19,000 The capability of developing a volta version for BTM miner is one thing, more important concern for us is that it’s just not profitable for us to put time into this development. Considering there are multiple algos, like cuck(at,ar)oo cycles and upcoming progpow, which are adopted by cryptocurrencires with much bigger marketcap, developing a BTM miner for a low user volumn market dose not make any sense. |
|
|
|
Volta's tensor cores don't support the INT8 mode supported by Turing's tensor cores, and BTM's tensority algo uses 8-bit matrix multiplication.
That being said, the @NebuTech devs could make a version specifically for Volta's tensor cores which used the FP16 matrix multiply mode supported by Volta, since you can encode INT8 numbers in FP16 format with no loss of precision.
yes, but there are other limitation other than operator type. |
|
|
|
|
Show Posts
